Posted On: 2025-04-07
Randomness in games is often implemented in ways that would befuddle a statistician. Whether it's a dealer at a casino adding extra cards to the deck (to prevent card-counting) or a digital game intentionally misrepresenting its odds (to make its players happier), there's always more to the way games use randomness than mere probability. For today's post, I'll be exploring two twists on randomness that are often used in games.
The primary function of randomness is unpredictability, but that unpredictability is often in service of something else - such as creating unique starting conditions (ie. shuffling cards before drawing a hand). Part of design is deciding the odds for different outcomes (ie. adding an extra 2 into a deck of cards to make it more likely), but digital games afford designers the ability to manipulate those odds at the moment that they are used. This allows the designer to use randomness where it's useful, but skew it in ways that serve the larger goal.
To explain through a concrete example - consider replacing tiles that were just removed from a game board. The general goal of using randomness here is to create a new (hopefully interesting) board for the player to continue playing on. Picking purely randomly, however, there is a (small) chance of putting tiles onto the board such that it looks exactly the same as before (ie. removing a red tile and replacing it with a [randomly-chosen] red tile.) The designer can avoid this by skewing the probabilities (ie. when removing a lot of red tiles, set the chance of drawing a red tile to be much lower than normal.)
While this simple example may seem straightforward, biased randomization is generally used in much more subtle ways, as small changes in randomness often have unforeseen consequences. Designers that use this approach often combine it with extensive testing and/or large amounts of player data, to validate that it's producing the right outcomes. Where it might be difficult to predict the effects of a small change (ie. going from 10% to 11% chance of drawing a red tile on the third turn), a designer running thousands of automated tests (Monte Carlo simulations) can see trends in important factors (ie. average number of turns until a win) and adjust their design accordingly.
Sometimes a game's design calls for deterministic randomness - where the "random" game elements produce consistent, reproducible outcomes. If that seems paradoxical, consider an analogy: a group of card players give their deck to a designated "shuffler", who promises to randomly arrange the cards in it. After receiving the deck, the shuffler (privately) shuffles it and then looks through the deck and writes down the order of the cards. The shuffler then returns the deck, and the group of players enjoy a (truly random) game. Later, another group gives their deck to the shuffler. This time, instead of shuffling, the shuffler looks through the deck and arranges the cards so that they are in the order written down from the first group. Then the shuffler gives the deck back to this second group of players, and they enjoy a (deterministic random) game: the cards are all drawn in the exact same order as the first group of players, but from the point of view of the second group of players, it still looks "random".
Deterministic randomness is useful for a wide range of situations - some of which are more visible to players than others. Using it for the initial setup (like the card game example above) allows a player to choose to retry a game using the exact same start (or share that start with others.) Using it in multiplayer can keep every player "synchronized" - everyone sees the same outcome for every random event*. In a game where players can "undo" an action, deterministic randomness makes sure that undoing and then repeating an action always produces the same outcome**.
Both kinds of random are (at this point) extremely common in games. Biased random is an immensely valuable tool to optimize outcomes in cutthroat markets (ie. mobile games), and a quick way to smooth over rough edges in design (ie. changing probability to interrupt a run of bad luck.) Deterministic random - which has been around in many forms for a long time - has grown alongside interest in genres that favor random starting conditions (survival/rogue-likes/etc.) and streaming/simultaneous play (where sharing the "seed" for a particular start is a common social activity.)
Hopefully this exploration of different types of randomness has been interesting for you. As always, if you have any thoughts or feedback, please let me know.